Pre-async

Work on SharedTensor

* Refactor Shape struct
* Implement realloc method
* Add write to memory and write to memory offset methods to SharedTensor
* Refactor native memory representation